De 20 a 25 años:
8.3 Comparación de los niveles de consumo y de la composición de los alimentos
A few words on notation and terminology for what follows …rst: will be taken to mean in what follows all the variables other than the ones being generated in the particular block under consid-eration. Furthermore, f f; f f ; X will denote the entire vector for the dependent variable, the latent dependent variable, and the entire matrix of explanatory variables, respectively, (periods 1; :::; T ), and f ft; f ft; Xt will denote the dependent variable, the latent dependent variable, and the explanatory variables for period t, respectively. The word "conditional" will be taken to mean conditional on everything, except of course for the variable(s) being generated in the particular block under consideration.
Algorithm for the Benchmark Model:
Generating the Variance: Inverted gamma distributions are convenient priors for the variance, since when multiplied by the conditional likelihood, they result in conditional posteriors which are also inverted gammas54, that we know how to sample from:
So, if the prior for 2 is IG( 20; 20), where IG stands for inverted gamma, then the conditional posterior is also IG( 21; 21), where 1= 0+ T , and 1= 0+ " 0" , where " is the T 1 vector of latent error terms of equation (2).
I conducted experiments using both ‡at priors55 and various forms of inverted gamma priors (I let each of the two parameters of the Gamma pdf vary from 0.1 to 10 and I tried various combinations of the two parameters within that range). The results were quite similar in all these experiments.
Generating the coe¢ cients of the explanatory variables: A ‡at prior for this block results in a Gaussian conditional posterior from which I can sample easily: In particular, this conditional
5 4See, for instance, Kim & Nelson (1999) for the derivation of this.
5 5An appropriate ‡at prior for the variance is the positive half of the real line and it results in a posterior for the variance which is proportional to the conditional likelihood for the variance (thus it’s proportional to ( 2) T2 e 2 21 " 0" ) and whose support is the positive real line. This is also of the Gamma form.
posterior is, in a standard way, N ((X0X) 1X0f f ; 2(X0X) 1). For a derivation of this, see, for instance, Albert and Chib (1993b). I also conducted robustness checks using proper conjugate Gaussian priors, and the results did not change signi…cantly.
The required stationarity constraints on the coe¢ cients of the lags of the latent dependent vari-able are implemented with rejection sampling, whereby draws from the posterior for the coe¢ cients are taken until the constraints are satis…ed, (and the draws are discarded when they do not satisfy the constraints).
Generating the latent dependent variables: I use a single-move smoothing algorithm here, which entails simulating each f ft, t = 1; :::; T , one by one in separate blocks, while also conditioning on all the data, and all the other parameters, including all the other latent variables, for each block.
The algorithm is derived as follows:
Let g(f ftj ; ff; X) denote the conditional distribution of fft, and let ff ft denote all the latent variables for periods 1; :::; t, and let ff f6=t denote all the latent variables for all periods except for t, and similarly let ff ft denote all the dependent variables for periods 1; :::; t. The dependence on the parameters other than the latent variables and on the explanatory variables is suppressed in what follows for convenience. Furthermore, for expositional purposes, I present the case of one lag for the latent variable. The proof for more than one lags is the same. So, we have that:
g(f ftjff f6=t; ff fT) = g(f ftjff f6=t; ff ft; f ft+1; :::; f fT)
= g(f ft; f ft+1; :::; f fTjff f6=t; ff ft) g(f ft+1; :::; f fTjff f6=t; ff ft)
= g(f ftjff f6=t; ff ft)g(f ft+1; :::; f fTjff f6=t; ff ft; f ft) g(f ft+1; :::; f fTjff f6=t; ff ft)
= g(f ftjff f6=t; ff ft)
= g(f ftjff ft 1; f ft+1; :::; f fT; ff ft 1; f ft)
= g(f ft; f ft+2; :::; f fTjff ft 1; f ft+1; ff ft 1; f ft) g(f ft+2; :::; f fTjff ft 1; f ft+1; ff ft 1; f ft) / g(fft; f ft+2; :::; f fTjff ft 1; f ft+1; ff ft 1; f ft)
= g(f ftjff ft 1; f ft+1; ff ft 1; f ft)g(f ft+2; :::; f fTjff ft 1; f ft; f ft+1; ff ft 1; f ft) / g(fftjff ft 1; f ft+1; ff ft 1; f ft)
= g(f ftjfft 1; f ft+1; ff ft):
Note that the transition from the 3rd line to the 4th line is valid as f ft+1; :::; f fT do not depend on f ft, given ff f6=t. Note also that the transition from the 5th line to the 6th line is valid (for the case of models with one lag for the latent variable) because the denominator of the fraction of the 5th line does not depend on f ft.
The pdf of the resulting distribution, namely g(f ftjfft 1; f ft+1; ff ft) can be obtained from the joint distribution of all the error terms where f ft appears. For the case with one latent lag, f ft appears in the equations giving "t, and "t+1. The joint pdf of the error terms is Gaussian, and ignoring for a moment the e¤ect of conditioning on ff ft, it is easy to show56 that f ft is distributed (given f ft 1; f ft+1) as N ( 0Xt+ 1f ft 1+1+1(f f2 t+1 0Xt+1)
1
; 2), where ; Xt; Xt+1 are de…ned here to exclude the latent lag and its coe¢ cient, and 1 is the coe¢ cient of the latent lag.
The e¤ect of conditioning on ff ft is a truncation, and the form of the truncation is determined
5 6Just rewrite, in that joint pdf, each of the error terms as an expression of the latent variables that appear in the latent equation that corresponds to that error term.
by equation (3):
If f ft2 category j, then fft 2 (fft 1+ cj 1; f ft 1+ cj); 8j; 8t:
Thus, the required sampling task is that of sampling from a (univariate) truncated normal. The best way of doing that is a combination of sampling from a uniform and inverting the truncated normal cdf. Speci…cally, I wish to simulate the latent dependent variable, which has a Normal cdf F with mean and variance 2, but that is truncated between a and b. Let Z unif orm(0; 1): Then F 1(Z) F . Therefore, and since F (f ft) = (f ft ) (
a )
(b ) (a ) , where is the standard normal cdf, I simulate the latent dependent variable by sampling from: 1fZ[ (b ) (a )]+ (a )g+ .
Additional blocks needed for the threshold coe¢ cients:
Here I can adopt a di¤use prior for the threshold coe¢ cients as then the conditional posterior becomes a uniform distribution (that we know how to sample from). In particular, it can be shown (see Albert and Chib (1993b) ) that the conditional posterior for cj; j = 1; :::; J , (where the conditioning is on the other threshold coe¢ cients too, in addition to everything else) is:
unif orm[maxfmaxt ffft : f ft= jg; cj 1g; minfmint ffft : f ft= j + 1g; cj+1g].
Additional blocks needed to implement the regime-switching extension:
Generating the State Variables: The states, S1; :::; ST are Bernoulli random variables. Let SfT = [S1:::ST]: Let Y1; :::; YT denote the "data" (that is, the dependent variable, the latent depen-dent variable, and the explanatory variables) for periods 1; :::; T , respectively, and let eYt denote all the data up to period t; t = 1; :::; T . Let g( fSTj ; fYT) denote the conditional distribution of fST. Following Kim and Nelson (1999), I adopt the Multi-Move Gibbs sampling approach; that is, I draw all of the states together in a single block, that is, from g( fSTj ; fYT), which, as Kim and Nelson (1999) demonstrate can be simpli…ed as follows:
g( fSTj ; fYT) = g(S1; :::; STj ; fYT)
= g(STj ; fYT)g(ST 1; :::; S1jST; ; fYT)
= g(STj ; fYT)g(ST 1jST; ; fYT)g(ST 2; :::; S1jST; ST 1; ; fYT)
= g(STj ; fYT)g(ST 1jST; ; fYT)g(ST 2jST; ST 1; ; fYT):::g(S1jST 1; :::; S2; ; fYT)
= g(STj ; fYT)g(ST 1jST; ; ]YT 1)g(ST 2jST 1; ; ]YT 2):::g(S1jS2; Y1)
= g(STj ; fYT)YT 1
t=1 g(StjSt+1; ; eYt):
The simpli…cations above occur because of the Markov property of the states. So, and as suggested by the last expression, we can generate …rst ST conditional on ; fYT, and then for t = T 1; :::; 1, we generate Stconditional on eYtand St+1. So, I run Hamilton’s …lter (Hamilton (1989)) to obtain, for all t; g(Stj ; eYt), and g(STj ; fYT) in particular, from which I generate ST. Then I generate the states of the previous periods using the following:
g(Stj ; eYt; St+1) = g(St;St+1j ; eYt)
g(St+1j ; eYt) = g(St+1jSt; ; eYt)g(Stj ; eYt)
g(St+1j ; eYt) :57
Then, using this expression, I calculate Pr[St= 1j ; St+1; eYt] as follows:
Pr[St= 1j ; St+1; eYt] = P1g(St+1jSt=1; )g(St=1j ; eYt) j=0g(St+1jSt=j; )g(St=jj ; eYt).
To determine whether St is then 0 or 1, I take a draw from the uniform distribution between 0 and 1, and if the generated number is less than or equal to this probability I set St= 1. Otherwise I set St= 0:
Generating the Transition Probabilities: The beta distribution (used here) is a convenient conjugate prior for the transition probabilities, as it is easy to sample from a beta distribution. The priors for q and p are: q~beta(u00; u01); p~beta(u11; u10),
where I set the hyperparameters of the priors equal to di¤erent values and I observe that the results are quite similar across these experiments.
The conditional likelihood is: L(q; p) = qn00(1 q)n01pn11(1 p)n10;
where nij is equal to the number of transitions from state i to state j, i; j = 0; 1:
The resulting posteriors are also beta distributions58: q~beta(u00+ n00; u01+ n01); p~beta(u11+ n11; u10+ n10).
5 7g(St+1jSt) is the transition probability (generated in separate blocks), and g(Stj ; eYt); g(St+1j ; eYt) have been obtained from Hamilton’s …lter.
5 8See Kim and Nelson (1999) for a derivation of this result.
Generating the Variances: As with the benchmark model, the priors used here are inverted gammas:
To generate 20 conditional on h1, I divide both sides of the latent equation by p
1 + h1St, and it can be shown that if the conditional prior for 20 is IG(20; 20), then the conditional posterior is also IG(21; 21), where 1 = 0+ T , and 1 = 0+PT
t=1(residuals )2, where residuals are equal to the normalized LHS of the latent equation minus the normalized RHS of the latent equation.
To generate h1 (and thus 21) conditional on 20, I divide both sides of the latent equation by 0, and, as before, it can be shown that if the conditional prior for h1 is IG(23; 23), then the conditional posterior is also IG(24; 24), where 4= 3+ T , and 4 = 3+PT
t=1(residuals )2, where residuals are as before (where now the division is done by 0), and where T is equal to the number of periods during which we are at the high volatility state.
As with the single variance case, I conducted experiments using both ‡at priors and various forms of inverted gamma priors (I let each of the two parameters of the Gamma pdf vary from 0.1 to 10 and I tried various combinations of the two parameters within that range). The results were quite similar in all these experiments.
The normalization constraint that h1 > 0, which means that 21 is constrained to be the high state variance and thus that it must be higher than 20 (that is constrained to be the low state variance), is implemented with rejection sampling.
Additional blocks needed for the Time Varying Parameters:
Generating the Time Varying Parameters: In contrast to the case of the latent variables, obtaining the smoothing algorithm for the Time Varying Parameters is standard because here we can usefully employ a state-space representation, with the Measurement Equation being the latent equation (equation (2000)), and with the Transition Equation being the driftless random walk for the TVP’s (equation (8)), together with the Kalman …lter.
Speci…cally, let eT = [ 1::: T]0. Let Y1; :::; YT denote the "data" (that is, the dependent variable, the latent dependent variable, and the explanatory variables) for periods 1; :::; T , respectively, and
let eYt denote all the data up to period t; t = 1; :::; T . Let g(eTj ; fYT) denote the conditional distribution of eT. Then, following Kim and Nelson (1999) I employ a multimove Gibbs-sampling approach, thus generating the entire eT as a block from its conditional distribution, g(eTj ; fYT).
The Markov property of the t’s ensure that convenient simpli…cations occur in g(eTj ; fYT), and in particular:
g(eTj ; fYT) = g( Tj ; fYT)g(eT 1j ; T; fYT)
= g( Tj ; fYT)g( T 1j ; T; fYT)g(eT 2j ; T 1; T; fYT)
= :::
= g( Tj ; fYT)g( T 1j ; T; fYT)g( T 2j ; T 1; fYT):::g( 1j ; 2; fYT)
= g( Tj ; fYT)g( T 1j ; T; ]YT 1)g(eT 2j ; T 1; ]YT 2):::g( 1j ; 2; fY1)
= g( Tj ; fYT)
TY1 t=1
g( tj ; t+1; eYt)
As suggested by this last expression, I …rst need to generate T from g( Tj ; fYT), and then, given t+1, generate t from g( tj ; t+1; eYt); t =; :::; T 1. Thus, I …rst generate T from g( Tj ; fYT)~N ( T jT; PT jT), and then t; for t = T 1; ::; 1 from g( tj ; t+1; eYt)~N ( tjt; t+1; Ptjt; t+1), where T jT = E( Tj ; fYT); PT jT = Cov( Tj ; fYT); tjt; t+1 = E( tj ; eYt; t+1) = E( tj ; tjt; t+1);
Ptjt; t+1 = Cov( Tj ; fYT; t+1) = Cov( Tj ; tjt; t+1). The updating terms T jT; PT jT; (and also all tjt; Ptjt; t = 1; :::; T ) can be derived in a standard way using the Kalman …lter59. The same holds true for the terms tjt; t+1, and Ptjt; t+1 since they can also be viewed as updating terms in which the updating is done not with Yt, but with t+1, which has been generated, and thus can be considered as observed data.
The initial values, 0j0 are arbitrary, with P0j0 having large diagonal elements (so that large uncertainty is attached to 0j0).
The re‡ecting barriers imposing the stability condition on the coe¢ cients of the lags of the dependent variable are implemented with rejection sampling, done separately for each time period t = 1; :::; T:
5 9See Hamilton (1994), and Kim and Nelson (1999).
Generating the precision matrix H: The prior for H is Wishart, W ( 0; H0), where I set
0 = 0; H0 1 = 0, and then the conditional posterior for H is also Wishart, W ( 1; H1), where
1 = T + 0; H1 = [H01+ XT
t=1
( t t 1)( t t 1)0] 1.
A note on the computer code: All of the computer code for this paper was written in Gauss, Version 3.2.34. The seed was always …xed at 180303.
References
Albert, J., and S. Chib (1993a): "Bayes Inference via Gibbs Sampling of Autoregressive Time Series Subject to Markov Mean and Variance Shifts," Journal of Business and Economic Statistics, January, Vol., 11, No. 1, pp. 1-15.
Albert, J., and S. Chib (1993b): "Bayesian Analysis of Binary and Polychotomous Response Data,"
Journal of the American Statistical Association, June, Vol., 88, No. 422, pp. 669-679.
Bernanke, B. S., and A. S. Blinder (1992): “The Federal Funds Rate and the Channels of Monetary Transmission,” American Economic Review, 82, 901-21.
Bernanke, B., and I. Mihov (1998): "Measuring Monetary Policy," Quarterly Journal of Economics 113(3), 869-902.
Bernstein, S. (1917): Theory of Probability (in Russian), 4th Edition (1946) Gostekhizdat, Moscow-Leningrad.
Blanchard, O.J., and J. Simon (2001): "The Long and Large Decline in U.S. Output Volatility,"
Brookings Papers on Economic Activity, 2001-1, 135-174.
Boivin, J. (2004): "Has U.S. Monetary Policy Changed? Evidence from Drifting Coe¢ cients and Real-Time Data," mimeo, Columbia.
Chernozhukov, V., and H. Hong (2003): "An MCMC approach to classical estimation," Journal of Econometrics (115), 293-346.
Chib, S. (2001): "Markov Chain Monte Carlo Methods: Computation and Inference." In: Heckman, J, and Leamer, E. (Eds.), Handbook of Econometrics, Vol. 5, North-Holland, Amsterdam, p. 3569-3649.
Clarida, R., Galí, J., and M. Gertler (1998): "Monetary Policy Rules in Practice: Some International Evidence," European Economic Review 42: 1033-1067.
Clarida, R., Galí, J., and M. Gertler (1999): "The Science of Monetary Policy: A New Keynesian Perspective," Journal of Economic Literature Vol XXXVII, No. 4: 1661-1707.
Clarida, R., Galí, J., and M. Gertler (2000): "Monetary Policy Rules and Macroeconomic Stability:
Evidence and Some Theory," Quarterly Journal of Economics 115(1), 147-180.
Cogley, T., and T. Sargent (2001): "Evolving Post-World War II U.S. In‡ation Dynamics," NBER Macroeconomics Annual 16, pp. 331-373.
Cogley, T., and T. Sargent (2002): "Drifts and Volatilities: Monetary Policies and Outcomes in the Post WWII U.S.," mimeo, New York University.
Cook, T., and T. Hahn (1989): "The E¤ect of Changes in the Federal Funds Rate Target on Market Interest Rates in the 1970’s," Journal of Monetary Economics 24: 331-351.
Dueker, Michael (1999a): "Measuring Monetary Policy Inertia in Target Fed Funds Rate Changes,"
Federal Reserve Bank of St. Louis Review 81(5): 3-9.
Dueker, Michael (1999b): "Conditional Heteroskedasticity in Qualitative Response Models of Time Series: A Gibbs Sampling Approach to the Bank Prime Rate," Journal of Business and Economic Statistics 17(4): 466-472.
Dueker, M., and K. Wesche (2003): "European Business Cycles: New Indices and their Synchronic-ity", Economic Inquiry 41(1): 116-131.
Eichengreen, B., M. Watson, and R. Grossman (1985): "Bank Rate Policy under the Interwar Gold Standard: A Dynamic Probit Model," Economic Journal 95: 725-745.
Geweke, John (1991): “E¢ cient Simulation from the Multivariate Normal and Student Distributions subject to Linear Constraints,” Computer Science and Statistics: Proceedings of the Twenty-Third Symposium on the Interface, 571-578.
Geweke, J., and M. Keane (2001): "Computationally Intensive Methods for Integration in Econo-metrics." In: Heckman, J, and Leamer, E. (Eds.), Handbook of Econometrics, Vol. 5, North-Holland, Amsterdam, p. 3463-3568.
Goodfriend, M. (1991): "Interest Rates and the Conduct of Monetary Policy," Carnegie-Rochester Conference Series on Public Policy, XXXIV, 7-30.
Hajivassiliou, Vassilis (1990): “Smooth Simulation Estimation of Panel Data LDV Models”, unpub-lished.
Hajivassiliou, V., and D. McFadden (1998): “The Method of Simulated Scores for the Estimation of LDV Models,” Econometrica 66(4), 863-96.
Hamilton, James D. (1989): "A New Approach to the Economic Analysis of Nonstationary Time Series and the Business Cycle," Econometrica, 57(2), 357-384.
Hamilton, James D. (1994): Time Series Analysis, Princeton University Press, Princeton.
Hamilton, J., and O. Jordà (2002): “A Model for the Federal Funds Rate Target,” forthcoming, Journal of Political Economy.
Hamilton, J., Waggoner, D., and T. Zha (2003): "Normalization in Econometrics," manuscript, UCSD.
Jalil, M. (2004): "Monetary Policy in retrospective: a Taylor rule inspired exercise," mimeo, UCSD.
Judd, J., and G. Rudebusch (1998): "Taylor’s Rule and the Fed: 1970-1997," Federal Reserve Bank of San Francisco Economic Review Number 3, 3-16.
Keane, Michael (1994): “A computationally Practical Simulation Estimator for Panel Data,”Econo-metrica, 62(1), 95-116.
Kim, C., and C. Nelson (1999a): State-Space Models With Regime Switching, MIT Press, Cambridge.
Kim, C., and C. Nelson (1999b): "Has the U.S. Economy Become More Stable? A Bayesian Approach Based on a Markov-Switching Model of Business Cycle," The Review of Economics and Statistics, 81, 608-616.
Lee, Lung-fei (1999): "Estimation of Dynamic and ARCH Tobit Models", Journal of Econometrics 92: 355-390.
Lehmann E. L., and G. Casella (1998): Theory of Point Estimation, 2nd edition, Springer.
Lerman, S., and C. Manski (1981): “On the Use of Simulated Frequencies to Approximate Choice Probabilities.”In: Manski, C., and McFadden, D. (Eds.), Structural Analysis of Discrete Data with Econometric Applications, MIT Press, Cambridge, p. 305-319.
Mavroeidis, S. (2002): "Identi…cation and Mispeci…cation Issues in Forward-Looking Models," UVA Econometrics Discussion Paper 2002/21.
Mavroeidis, S. (2004): "Weak Identi…cation of Forward-looking Models in Monetary Economics,"
Oxford Bulletin of Economics and Statistics, 66, supplement: 609-635.
McCulloch, R., and P. Rossi (1994): "An Exact Likelihood Analysis of the Multinomial Probit Model," Journal of Econometrics 64: 207-240.
McFadden, D. (1989): “A Method of Simulated Moments for Estimation of Discrete Choice Models without Numerical Integration,” Econometrica 57, 995-1026.
Meulendyke, Ann Marie (1998): U.S. Monetary Policy and Financial Markets. New York: Federal Reserve Bank of New York.
Monokroussos, G. (2004): "A Classical MCMC Approach to the Estimation of Limited Dependent Variable Models of Time Series," mimeo, UCSD.
Orphanides, A. (2001): "Monetary Policy Rules Based on Real-Time Data," American Economic Review 91(4), 964-985.
Orphanides, A. (2002): "Monetary Policy Rules and the Great In‡ation," American Economic Review, Papers and Proceedings, 92(2), 115-120, May 2002.
Orphanides, A. (2004): "Monetary Policy Rules, Macroeconomic Stability and In‡ation: A View from the Trenches," Journal of Money, Credit and Banking, 35(6).
Pakes, A., and D. Pollard (1989): "Simulation and the asymptotics of optimization estimators,"
Econometrica, 57, 1027-1057.
Piazzesi, Monika (2001): "An Econometric Model of the Yield Curve with Macroeconomic Jump E¤ects," NBER Working Paper No. 8246.
Primiceri, G. (2004): "Why In‡ation Rose and Fell: Policymakers’Beliefs and US Postwar Stabi-lization Policy," mimeo, Princeton.
Robert, C.P., and G. Casella (1999): Monte Carlo Statistical Methods, Springer.
Rotemberg, J., and M. Woodford (1997): "An Optimization-Based Econometric Framework for the Evaluation of Monetary Policy," NBER Macroeconomics Annual, Bernanke, B., and J. Rotemberg, eds.
Rotemberg, J., and M. Woodford (1999): "Interest Rate Rules in an Estimated Sticky Price Model,"
in Monetary Policy Rules, J. B. Taylor, ed.
Rudebusch, G. (1995): "Federal Reserve Interest Rate Targeting, Rational Expectations and the Term Structure," Journal of Monetary Economics 35: 245-274.
Sack, B., and W. Wieland (2000): "Interest Rate Smoothing and Optimal Monetary Policy: A Review of Recent Empirical Evidence," Journal of Economics and Business 52, 205-228.
Sims, C. (1999): "Drift and Breaks in Monetary Policy," mimeo, Princeton University.
Sims, C., and T. Zha (2004): "Were There Regime Switches in US Monetary Policy?," mimeo, Princeton University.
Stock, J. H., and M. W. Watson (2003): "Understanding Changes in International Business Cycle Dynamics," mimeo, Princeton University.
Taylor, John B. (1993): “Discretion Versus Policy Rules in Practice,” Carnegie-Rochester Confer-ence Series on Public Policy 39 (December 1993), 195-214.
Taylor, J. B. (1999): Monetary Policy Rules, The University of Chicago Press.
Tanner, M.A., and W.H. Wong (1987): "The calculation of posterior distributions by data augmen-tation," Journal of the Americal Statistical Association, 82: 528-549.
von Mises, R. (1931), Wahrscheinlichkeitsrechnung, Springer-Verlag, Berlin.
Note: Vertical line represents Paul Volcker's appointment as Chairman of the Federal Reserve in August of 1979.
Figure 1 U.S. Inflation
0 2 4 6 8 10 12
1965 1969 1973 1977 1981 1985 1989 1993 1997
variable mean std. dev. median 2.5% qntl 5% qntl 10% qntl 90% qntl 95% qntl 97.5% qntl
intercept 0.0108 0.1555 0.0069 -0.2637 -0.2252 -0.1819 0.2037 0.2783 0.3406
variance 0.1593 0.0598 0.1466 0.0777 0.0836 0.0961 0.2408 0.2708 0.3061
1st lag 0.7544 0.0685 0.7627 0.5782 0.6292 0.6653 0.83 0.8475 0.8635
inflation 0.4808 0.1363 0.4648 0.2567 0.2838 0.3174 0.6544 0.7229 0.7952
output gap 0.4035 0.1006 0.3908 0.2401 0.2605 0.2879 0.5332 0.5947 0.647
intercept 0.0256 0.159 0.0207 -0.2641 -0.2129 -0.172 0.2224 0.312 0.3579
variance 0.1659 0.0638 0.1526 0.0777 0.0863 0.0974 0.2512 0.2837 0.3098
1st lag 0.7542 0.0668 0.7657 0.5903 0.6228 0.6686 0.8287 0.844 0.8648
inflation 0.4754 0.1325 0.4617 0.2532 0.283 0.3204 0.6433 0.7049 0.78
output gap 0.4042 0.1003 0.3892 0.2422 0.2601 0.2907 0.5363 0.5938 0.6421
intercept 0.0182 0.1538 0.0124 -0.2609 -0.2161 -0.17 0.2112 0.271 0.3358
variance 0.1588 0.057 0.1471 0.0798 0.0885 0.0986 0.2356 0.2694 0.3039
1st lag 0.7658 0.0663 0.7733 0.6099 0.6465 0.676 0.842 0.8593 0.8732
inflation 0.4502 0.1335 0.4392 0.2253 0.2531 0.2931 0.6268 0.6737 0.7421
output gap 0.3983 0.1001 0.3879 0.2354 0.2567 0.2817 0.5221 0.5858 0.6358
Note: The last 7 columns are quantiles of the posterior distibution.
Value of target at end of FOMC months Table 1
Greenspan Period
FOMC monthly averages of fed funds rate
FOMC monthly averages of target Figure 2
Fed Funds Rate and its Target, FOMC Months
0
Fed Funds Rate, FOMC Monthly Averages Target, FOMC Monthly Averages Target, end-of-FOMC month values
variable mean std. dev. median 2.5% qntl 5% qntl 10% qntl 90% qntl 95% qntl 97.5% qntl
intercept 0.9548 0.4496 0.9676 0.0755 0.2338 0.3988 1.5274 1.6921 1.8477
variance 1.1297 0.2374 1.0935 0.7517 0.7925 0.8518 1.4398 1.5615 1.6794
1st lag 0.8664 0.0489 0.8672 0.7623 0.7819 0.8027 0.9267 0.94 0.9641
inflation 0.0682 0.0808 0.0676 -0.0809 -0.0596 -0.033 0.1769 0.2038 0.2326
output gap 0.3006 0.1 0.3004 0.1111 0.1376 0.1729 0.4251 0.4666 0.4941
intercept 0.9451 0.4472 0.956 0.0531 0.2338 0.3838 1.4903 1.6664 1.8351
variance 1.1216 0.2364 1.0856 0.7403 0.7806 0.8314 1.4465 1.5747 1.6534
1st lag 0.8661 0.0474 0.8687 0.7623 0.7817 0.802 0.9261 0.9368 0.9526
inflation 0.0704 0.08 0.0691 -0.0716 -0.0484 -0.0272 0.1817 0.2045 0.2254
output gap 0.3014 0.0996 0.3009 0.1081 0.1394 0.1715 0.4317 0.4671 0.4906
intercept 0.9645 0.4522 0.973 0.082 0.2288 0.4021 1.5374 1.7037 1.9593
variance 1.1378 0.2383 1.1039 0.7705 0.814 0.8656 1.437 1.5324 1.7086
1st lag 0.8668 0.0505 0.8666 0.7656 0.7838 0.8027 0.9282 0.9466 0.971
inflation 0.066 0.0815 0.0636 -0.0837 -0.0658 -0.0344 0.165 0.1936 0.2362
output gap 0.2997 0.1004 0.2999 0.1127 0.1337 0.1742 0.4242 0.4647 0.4951
variable mean std. dev. median 2.5% qntl 5% qntl 10% qntl 90% qntl 95% qntl 97.5% qntl
intercept -0.0439 0.279 -0.0281 -0.5669 -0.5025 -0.4088 0.3199 0.401 0.4955
variance 1.2317 0.196 1.2084 0.902 0.943 0.9919 1.4954 1.5877 1.6677
1st lag 0.7679 0.0435 0.7668 0.6826 0.6958 0.7117 0.823 0.8396 0.8535
inflation 0.4518 0.0865 0.4555 0.2683 0.3054 0.343 0.5552 0.5882 0.6256
output gap 0.1144 0.0874 0.1133 -0.0532 -0.0301 -0.0007 0.23 0.265 0.2875
intercept -0.0378 0.2695 -0.0231 -0.5487 -0.4879 -0.3941 0.3163 0.4054 0.5055
variance 1.2171 0.1839 1.2005 0.8945 0.9412 0.9929 1.4652 1.5274 1.645
1st lag 0.7688 0.0424 0.7668 0.6882 0.6963 0.7165 0.8238 0.8408 0.8524
inflation 0.4487 0.0843 0.4532 0.27 0.3021 0.3393 0.5507 0.5825 0.6108
output gap 0.1115 0.0836 0.1119 -0.0532 -0.0353 0.0049 0.2224 0.2564 0.2908
intercept -0.05 0.2882 -0.0338 -0.5914 -0.5273 -0.4269 0.336 0.3983 0.4955
variance 1.2464 0.2066 1.225 0.9098 0.9516 0.9907 1.534 1.6198 1.6802
1st lag 0.767 0.0446 0.767 0.6745 0.6939 0.7103 0.8214 0.8369 0.8556
inflation 0.455 0.0886 0.4565 0.2683 0.3054 0.3482 0.561 0.5967 0.6458
output gap 0.1173 0.091 0.1138 -0.0523 -0.0296 -0.0034 0.2366 0.2695 0.2842
Note: Benchmark specification with 1 lag of the dependent variable.
The last 7 columns are quantiles of the posterior distibution.
1:500 iterations
intercept
intercept
variable mean std. dev. median 2.5% qntl 5% qntl 10% qntl 90% qntl 95% qntl 97.5% qntl
intercept 0.9356 0.4392 0.9186 0.0721 0.2163 0.3811 1.4958 1.648 1.8073
variance 0.9956 0.2023 0.9776 0.6604 0.7078 0.7527 1.2621 1.3614 1.4329
1st lag 0.7471 0.1324 0.7437 0.5002 0.5298 0.5759 0.9192 0.9637 1.0185
2nd lag 0.1446 0.1315 0.1465 -0.1242 -0.0692 -0.0224 0.3136 0.3567 0.3852
inflation 0.0426 0.076 0.0407 -0.0988 -0.0825 -0.055 0.1366 0.1771 0.1975
output gap 0.3246 0.1004 0.3236 0.1236 0.1555 0.1917 0.4485 0.4899 0.5273
intercept 0.9581 0.4305 0.9564 0.1795 0.2535 0.4299 1.5292 1.6498 1.8073
variance 0.9812 0.1959 0.9586 0.6543 0.7013 0.7527 1.2472 1.3257 1.4036
1st lag 0.7505 0.1275 0.7524 0.498 0.5424 0.5749 0.9109 0.9594 1.0025
2nd lag 0.1428 0.1292 0.1446 -0.1025 -0.0692 -0.0203 0.3151 0.3574 0.3803
inflation 0.0368 0.0753 0.0358 -0.1053 -0.0866 -0.0623 0.1335 0.1755 0.1894
output gap 0.3283 0.0998 0.3295 0.1211 0.1555 0.1994 0.448 0.4899 0.5178
intercept 0.9131 0.447 0.9114 0.0261 0.1619 0.348 1.4582 1.6307 1.8641
variance 1.0101 0.2078 1.0045 0.6685 0.7117 0.7527 1.2729 1.3871 1.4782
1st lag 0.7436 0.1372 0.7376 0.5015 0.5244 0.5759 0.9251 0.969 1.0376
2nd lag 0.1465 0.1338 0.1473 -0.1264 -0.0701 -0.0259 0.3131 0.3563 0.3979
inflation 0.0485 0.0764 0.0483 -0.0922 -0.0752 -0.0501 0.1414 0.1794 0.2163
output gap 0.3209 0.1011 0.3174 0.1316 0.1521 0.1857 0.452 0.4899 0.5291
variable mean std. dev. median 2.5% qntl 5% qntl 10% qntl 90% qntl 95% qntl 97.5% qntl
intercept -0.0455 0.2833 -0.0488 -0.5756 -0.4926 -0.41 0.3324 0.447 0.5006
variance 1.2293 0.2007 1.2099 0.8723 0.922 0.9869 1.4853 1.5926 1.6777
1st lag 0.7359 0.1059 0.7381 0.5237 0.5634 0.6067 0.8689 0.9145 0.9506
2nd lag 0.0299 0.0975 0.0318 -0.1635 -0.1332 -0.0927 0.1509 0.1866 0.2211
inflation 0.4595 0.0888 0.4606 0.276 0.3121 0.3479 0.5729 0.611 0.6369
output gap 0.1131 0.0872 0.1112 -0.0712 -0.037 -0.0022 0.2261 0.2586 0.2867
intercept -0.0499 0.2899 -0.0472 -0.6045 -0.5237 -0.4235 0.3301 0.4455 0.4845
variance 1.2242 0.2008 1.2044 0.8587 0.9074 0.9731 1.4687 1.5845 1.6803
1st lag 0.731 0.1102 0.73 0.5068 0.5578 0.6003 0.8696 0.9218 0.9608
2nd lag 0.0359 0.1034 0.0453 -0.1707 -0.1484 -0.1101 0.1613 0.1958 0.2358
inflation 0.4596 0.0883 0.4623 0.2659 0.3075 0.3479 0.568 0.6061 0.6236
output gap 0.1157 0.0893 0.1137 -0.0722 -0.0343 0.0026 0.2343 0.2632 0.2878
intercept -0.0412 0.2767 -0.057 -0.5244 -0.4727 -0.398 0.3324 0.447 0.5662
variance 1.2344 0.2007 1.2181 0.8935 0.9291 1.0002 1.4963 1.5955 1.6705
1st lag 0.7409 0.1012 0.7468 0.5495 0.5791 0.6097 0.8688 0.9091 0.9421
2nd lag 0.0238 0.0909 0.023 -0.1462 -0.1221 -0.0856 0.1384 0.1713 0.2024
inflation 0.4595 0.0894 0.4559 0.2963 0.3183 0.347 0.5765 0.6129 0.6407
output gap 0.1105 0.085 0.11 -0.0706 -0.0387 -0.0067 0.2195 0.2484 0.276
Note: Benchmark specification with 2 lags of the dependent variable.
The last 7 columns are quantiles of the posterior distibution.
Table 3
variable mean std. dev. median 2.5% qntl 5% qntl 10% qntl 90% qntl 95% qntl 97.5% qntl
intercept 0.9719 0.4289 0.9627 0.1484 0.2933 0.4368 1.5111 1.6939 1.7968
variance 1.0825 0.211 1.0622 0.724 0.7798 0.83 1.3447 1.4446 1.5276
1st lag 0.806 0.0536 0.809 0.6962 0.71 0.7332 0.8717 0.89 0.9039
inflation 0.1586 0.0745 0.1581 0.0105 0.0332 0.0628 0.2531 0.2817 0.3002
output gap 0.3716 0.1047 0.3671 0.1773 0.2038 0.2365 0.5084 0.538 0.5776
intercept 1.1139 0.4062 1.1045 0.3724 0.4543 0.5789 1.6488 1.8077 1.9074
variance 0.9969 0.1949 0.9781 0.6548 0.7047 0.7624 1.2473 1.3441 1.408
1st lag 0.7766 0.0585 0.7791 0.6607 0.6763 0.6998 0.8505 0.8732 0.8905
inflation 0.168 0.0761 0.1683 0.0196 0.0435 0.0731 0.2649 0.2914 0.3118
output gap 0.4182 0.1086 0.418 0.2109 0.242 0.2792 0.5578 0.5896 0.6281
intercept 1.5299 0.4589 1.5261 0.654 0.8093 0.9675 2.1268 2.3443 2.4859
variance 1.0032 0.2014 0.9876 0.6515 0.7065 0.7603 1.2656 1.3821 1.4436
1st lag 0.7423 0.0683 0.7433 0.6062 0.6241 0.6511 0.8274 0.8538 0.8688
inflation 0.1288 0.0661 0.1284 0.0039 0.0228 0.045 0.2129 0.242 0.2676
output gap 0.4764 0.1264 0.4736 0.2427 0.2738 0.3137 0.639 0.6919 0.7486
variable mean std. dev. median 2.5% qntl 5% qntl 10% qntl 90% qntl 95% qntl 97.5% qntl
intercept -0.0201 0.2829 -0.0067 -0.556 -0.4786 -0.3931 0.3465 0.4354 0.5371
variance 1.338 0.2114 1.3161 0.9824 1.0276 1.0798 1.6191 1.7157 1.8195
1st lag 0.803 0.0435 0.8021 0.7179 0.7305 0.7481 0.858 0.8748 0.8883
inflation 0.3649 0.085 0.3698 0.1767 0.221 0.2602 0.4681 0.4991 0.5335
output gap 0.1422 0.0914 0.1419 -0.0361 -0.0094 0.0208 0.264 0.297 0.3291
intercept 0.0486 0.271 0.063 -0.4593 -0.4037 -0.3106 0.3954 0.4901 0.5768
variance 1.3476 0.2123 1.3252 0.9863 1.0262 1.0867 1.6331 1.7336 1.8252
1st lag 0.809 0.0435 0.8093 0.7211 0.7343 0.7528 0.8624 0.8786 0.8908
inflation 0.3245 0.0777 0.3271 0.1593 0.1922 0.2266 0.4205 0.4517 0.4785
output gap 0.1714 0.0918 0.1721 -0.0088 0.0184 0.051 0.2898 0.3261 0.3604
intercept 0.1416 0.2544 0.1548 -0.3444 -0.2749 -0.1907 0.4715 0.5412 0.6403
variance 1.2766 0.2026 1.2505 0.9322 0.9828 1.0283 1.5496 1.6379 1.7368
1st lag 0.7931 0.0419 0.793 0.7085 0.7238 0.7406 0.8458 0.8598 0.8717
inflation 0.3322 0.0683 0.3354 0.1857 0.2174 0.2442 0.4167 0.4418 0.4621
output gap 0.1727 0.089 0.1698 0.0009 0.0236 0.0569 0.2868 0.3209 0.3485
Note: Alternative Forecast Horizons - Benchmark specification with 1 lag of the dependent variable.
The last 7 columns are quantiles of the posterior distibution.
0-steps ahead